EDA: Overview of Suicide Rate in a Particular Year

Step-by-Step Data Preparation

1. Installing and launching required R packages

pacman::p_load("patchwork", "tmap", "ExPanDaR", "kableExtra", "ggstatsplot", "plotly", "DT", "tidyverse")

2. Loading the data

suicidedata_eda_formap <- read_csv("data/suicidedata_eda_formap.csv", show_col_types = FALSE)

suicidedata_eda <- read_csv("data/suicidedata_eda.csv", show_col_types = FALSE)

3. Dataset overview

We will check the suicidedata_eda as suicidedata_eda_formap is meant to be combined with the sf file (world)

3.1 Missing Values

missing.values <- suicidedata_eda %>%
    gather(key = "key", value = "val") %>%
    mutate(is.missing = is.na(val)) %>%
    group_by(key, is.missing) %>%
    summarise(num.missing = n()) %>%
    filter(is.missing==T) %>%
    select(-is.missing) %>%
    arrange(desc(num.missing)) 
missing.values %>% kable()

key	num.missing
DN	18360
DR	18360
POP	18360
SN	18360
SP	18360

Visualising Missing Values

missing_values <- function(vari = "age_name"){
prepare_missing_values_graph(suicidedata_eda, ts_id = vari)
}

By country

missing_values("country")

By gender

missing_values("sex_name")

By age group

missing_values("age_name")

By year

missing_values("year")

Conclusion : The missing values are because of age-standardised (only for Suicide Rates)

3.2 Descriptive statistics

descr_stats <- function(gender = "Both", age = "All ages") {
  
  descr <- prepare_descriptive_table(suicidedata_eda %>%
                                       filter(sex_name == gender,
                                              age_name == age) %>%
                                       select(!c(6,8)) %>%
                                       rename("Number of suicide" = "SN",
                                              "Number of deaths" = "DN",
                                              "Share of deaths from suicide (%)" = "SP",
                                              "Suicide rate" = "SR",
                                              "Mortality rate" = "DR"))
descr$kable_ret  %>%
  kable_styling("condensed", full_width = F, position = "center")
}

descr_stats("Both", "All ages")

Descriptive Statistics

	N	Mean	Std. dev.	Min.	25 %	Median	75 %	Max.
Number of suicide	6,120	3,874.828	18,425.617	0.126	93.934	532.835	1,882.305	220,356.720
Number of deaths	6,120	252,265.821	920,824.636	10.886	7,358.955	51,487.069	169,570.300	10,659,491.040
Share of deaths from suicide (%)	6,120	1.394	1.067	0.105	0.691	1.118	1.787	11.812
Suicide rate	6,120	11.339	9.395	1.427	5.424	8.224	14.601	95.565
Mortality rate	6,120	847.052	354.531	148.258	611.956	783.986	1,019.318	10,705.808

3.3 Distribution

3.3.1 Histogram of multiple variables

plot_multi_distribution <- function(gender = "Both", age = "All ages") {
  
suicidedata_hist <- suicidedata_eda %>% 
  filter(sex_name == gender,
         age_name == age) %>%
  select(c(7,9,10,11,12)) %>%
  rename("Number of suicide" = "SN",
         "Number of deaths" = "DN",
         "Share of deaths from suicide (%)" = "SP",
         "Suicide rate" = "SR",
         "Mortality rate" = "DR")
  
plot <- suicidedata_hist %>% 
  gather() %>%
  ggplot(aes(value)) +
  facet_wrap( ~key, ncol=3, scales="free") +
  geom_histogram()

ggplotly(plot)
}

plot_multi_distribution("Both", "All ages")

It does not make sense to look at numbers due to different population sizes. Hence we will only look at Suicide Rate, Mortality Rate and Share of deaths from suicide (%) from now on

3.3.2 Histogram of individual variables

plot_distribution <- function(gender = "Both", age = "All ages", vari = SR, title = "Suicide Rate", binwidth = 2) {
  
suicidedata_hist <- suicidedata_eda %>% 
  filter(sex_name == gender,
         age_name == age)

suicidedata_summary <- suicidedata_hist %>%
  summarise(mean = round(mean({{vari}})),
            median = round(median({{vari}})),
            ymax = as.numeric(round((IQR({{vari}})*1.5) + quantile({{vari}},0.75))),
            ymin = as.integer(min({{vari}})),
            rangemax = as.integer(max({{vari}}))) 

#computing summary statistics of mean, median and lower and upper whiskers in boxplot
var_mean <- suicidedata_summary$mean
var_median <- suicidedata_summary$median
ymax <- suicidedata_summary$ymax
ymin <- suicidedata_summary$ymin
rangemax <- suicidedata_summary$rangemax

#plotting histogram
h1 <- ggplot(suicidedata_hist, aes(x = {{vari}})) + 
  geom_histogram(color="black", fill="azure4", binwidth = binwidth)  
  
h <- h1 + scale_x_continuous(limits = c(0,rangemax), labels = scales::comma) +
  labs(x = paste0(title,", ",gender,", ",age), y = "Count") +
  geom_vline(aes(xintercept = var_mean), col="darkblue", linewidth=1, linetype = "dashed") +
  annotate("text", x=var_mean*1.05, y=max(ggplot_build(h1)$data[[1]]$count)*1.2, label="Mean:", 
           size=4, color="darkblue", hjust = 0) +
  annotate("text", x=var_mean*1.05, y=max(ggplot_build(h1)$data[[1]]$count)*1.1, label=format(var_mean, big.mark = ","),
           size=4, color="darkblue", hjust = 0) +
  geom_vline(aes(xintercept = var_median), col="lightpink4", linewidth=1, linetype = "dashed") +
  annotate("text", x=var_median*0.95, y=max(ggplot_build(h1)$data[[1]]$count)*1.2, label="Median:", 
           size=4, color="lightpink4", hjust = 1) +
  annotate("text", x=var_median*0.95, y=max(ggplot_build(h1)$data[[1]]$count)*1.1, label=format(var_median, big.mark = ","),
           size=4, color="lightpink4", hjust = 1) +
  theme(axis.text.x = element_text(size=8))

#plotting boxplot
b <- ggplot(suicidedata_hist, aes(y = {{vari}})) + 
  geom_boxplot(outlier.colour="firebrick", outlier.shape=16,
               outlier.size=1, notch=FALSE) + 
  coord_flip() + labs(y = "", x = "") + 
  scale_y_continuous(limits = c(0,rangemax), labels = scales::comma) +
  theme(axis.text = element_blank(), axis.ticks = element_blank()) + 
  stat_boxplot(geom="errorbar", width=0.5) +
  annotate("text", x=0.35, y=ymax, label=format(ymax, big.mark = ","), 
           size=3, color="lightpink4") +
  annotate("text", x=0.35, y=ymin, label=format(ymin, big.mark = ","), 
           size=3, color="lightpink4")

#combining plots
suicide_distri <- b / h + plot_layout(heights = c(1, 4)) 

suicide_distri
}

plot_distribution("Both", "All ages", SR, "Suicide Rate", 2)

3.4 Outliers

outliers_table <- function(gender = "Both", age = "All ages", vari = "Suicide rate") {
  
  outliers <- prepare_ext_obs_table(suicidedata_eda %>%
                                       filter(sex_name == gender,
                                              age_name == age) %>%
                                      select(!c(2,3,4,5)) %>%
                                      rename("Share of deaths from suicide (%)" = "SP",
                                             "Suicide rate" = "SR",
                                             "Mortality rate" = "DR"),
                                    n = 10,
                                    cs_id = "country",
                                    ts_id = "year",
                                    var = vari)
    
outliers$kable_ret  %>%
  kable_styling("condensed", full_width = F, position = "center")
}

outliers_table("Both", "All ages", "Suicide rate")

country	year	Suicide rate
Greenland	1,993	95.565
Greenland	1,994	94.916
Greenland	1,995	94.541
Greenland	1,990	93.745
Greenland	1,996	93.518
Greenland	1,992	93.423
Greenland	1,991	93.390
Greenland	1,997	92.778
Greenland	1,998	90.744
Greenland	1,999	88.223
...	...	...
Jamaica	1,993	1.586
Sao Tome and Principe	1,994	1.527
Jamaica	1,992	1.526
Sao Tome and Principe	1,995	1.514
Sao Tome and Principe	1,993	1.514
Sao Tome and Principe	1,992	1.467
Sao Tome and Principe	1,991	1.446
Jamaica	1,990	1.439
Jamaica	1,991	1.431
Sao Tome and Principe	1,990	1.427

More outliers on the high-end as compared to the other side with Greenland dominating the outliers. There might be interesting pattern here, hence windsorising the data using treat_outliers() function might not be recommended

4. Comparing between countries and regions within a particular year

4.1 Plotting the choropleth map

4.1.1 Joining with world map (tmap object)

The object World is a spatial object of class sf from the sf package; it is a data.frame with a special column that contains a geometry for each row, in this case polygons

Reference - https://cran.r-project.org/web/packages/tmap/vignettes/tmap-getstarted.html

data("World")

Joining the two dataframes together

suicidedata_eda_map <- left_join(World, 
                                 suicidedata_eda_formap %>% mutate(across(where(is.numeric), round, 2)),
                          by = c("iso_a3" = "code")) %>%
  select(!c(2,3,4,6,7,8,9,10,11,12,13,14,15)) %>%
  mutate(area = as.numeric(str_remove(`area`, 
                                " \\[km\\^2\\]")), 
         .after = region) %>%
  na.omit()

4.1.2 Creating function to plot choropleth map

plot_map_eda <- function(year, metric = "SR", gender = "T", style = "jenks"){
  
metric_text = case_when(metric == "SR" ~ "Suicide rate",
                        metric == "SP" ~ "Share of deaths from suicide (%)",
                        metric == "SN" ~ "Number of suicide",
                        metric == "DN" ~ "Number of deaths",
                        metric == "DR" ~ "Mortality rate")

age = case_when(metric == "SR" ~ "AS",
                metric == "SP" ~ "All",
                metric == "SN" ~ "All",
                metric == "DN" ~ "All",
                metric == "DR" ~ "All")

gender_text = case_when(gender == "T" ~ "Total",
                        gender == "M" ~ "Male",
                        gender == "F" ~ "Female")
  
tmap_mode("view")

tm_shape(suicidedata_eda_map |> 
           filter(year == year))+
  tm_fill(paste0(metric,"_",age,"_",gender), 
          style = style, 
          palette="YlOrBr", 
          id = "country",
          title = paste0(metric_text, ", ",gender_text,", ", year),
          popup.vars = c(value = paste0(metric,"_",age,"_",gender))) +
  
  tm_borders(col = "grey20",
             alpha = 0.5) 
}

plot_map_eda(2016, "SR", "M", "jenks")

tmap_mode("plot")

4.2 Statistical testing on difference in suicide metrics between Regions

Note that it does not make sense to compare the statistical testing between countries within a particular year as there is only one metric per country. As such, statistical test will be done for a time period for a country (in EDA Time Series segment)

4.2.1.1 Normality assumption

Plotting histogram by regions

plot_multi_distribution_region <- function(year, gender = "Both", age = "All ages", vari = SR) {
  
suicidedata_hist <- suicidedata_eda %>% 
  filter(year == year,
         sex_name == gender,
         age_name == age) %>%
  select(c(3,7,8,9,10,11,12))
  
plot <- suicidedata_hist %>% 
  ggplot(aes(x = {{vari}})) +
  facet_wrap( ~region, ncol=4, scales="free") +
  geom_histogram()

ggplotly(plot)
}

plot_multi_distribution_region(2010, "Both", "All ages", SR)

Anderson-Darling Test by Region

adtest_region <- function(year, gender = "Both", age = "All ages", vari = "SR") {

suicidedata_ad <- suicidedata_eda %>% 
  filter(year == year,
         sex_name == gender,
         age_name == age) %>%
  select(c(3,7,8,9,10,11,12))
    
normaltestlist <- list()
for (i in unique(suicidedata_ad$region)){
  subdf <- subset(x = suicidedata_ad, subset=region==i)
  normaltestlist[[i]] <- nortest::ad.test(subdf[[vari]])
}

normaltest <- tibble(town = unique(suicidedata_ad$region),
                     p_value = unlist(lapply(normaltestlist, `[[`, 2))) 

names(normaltest)[2] <- paste0(vari, " p-value")

DT::datatable(normaltest, class= "compact")
}

adtest_region(2010, "Both", "All ages", "SR")

4.2.1.2 ANOVA Test

plot_ANOVA_region <- function(year, gender = "Both", age = "All ages", vari = SR, title = "Suicide rate", type = "np", conf = 0.95) {

text <- case_when(type == "p" ~ "Mean",
                  type == "np" ~ "Median" )
    
suicidedata_ANOVA <- suicidedata_eda %>% 
  filter(year == year,
         sex_name == gender,
         age_name == age) %>%
  select(c(3,7,8,9,10,11,12))

suicidedata_ANOVA %>% ggbetweenstats(x = region, y = {{vari}},
               xlab = "Region", ylab = title,
               type = type, pairwise.comparisons = T, pairwise.display = "ns", 
               mean.ci = T, p.adjust.method = "fdr",  conf.level = conf,
               title = paste0("Comparison of ",text," ",title, " (", gender, " gender, ", age, ") across Regions in ", year),
               subtitle = paste0("One-way ANOVA at ", conf*100, "% confidence level"),
               package = "ggthemes", palette = "Tableau_10")

}

plot_ANOVA_region(2010, "Both", "Age-standardized", SR, type = "np")